[PPDiffusers] implement save_to_hf_hub for pipelines, models, and Preprocessor

[sijunhe]

PR types

PR changes

Description

# feature extractor
from paddlenlp.transformers import CLIPFeatureExtractor
feature_processing_config = {
  "crop_size": 32,
  "do_center_crop": True,
  "do_convert_rgb": True,
  "do_normalize": True,
  "do_resize": True,
  "feature_extractor_type": "CLIPFeatureExtractor",
  "image_mean": [
    0.48145466,
    0.4578275,
    0.40821073
  ],
  "image_std": [
    0.26862954,
    0.26130258,
    0.27577711
  ],
  "resample": 3,
  "size": 224
}
feature_extractor = CLIPFeatureExtractor(**feature_processing_config)
# scheduler
from ppdiffusers import DDIMScheduler
scheduler_config = {
  "beta_end": 0.012,
  "beta_schedule": "scaled_linear",
  "beta_start": 0.00085,
  "num_train_timesteps": 1000,
  "steps_offset": 1,
  "clip_sample": False
}
scheduler = DDIMScheduler(**scheduler_config)
# clip
from paddlenlp.transformers import CLIPTextConfig, CLIPTextModel, CLIPTokenizer
clip_text_config = CLIPTextConfig(**{
  "architectures": [
    "CLIPTextModel"
  ],
  "hidden_size": 32,
  "intermediate_size": 37,
  "max_position_embeddings": 77,
  "model_type": "clip_text_model",
  "num_attention_heads": 4,
  "num_hidden_layers": 5,
  "vocab_size": 1000
})
text_encoder = CLIPTextModel(clip_text_config)
tokenizer = CLIPTokenizer.from_pretrained("test_clip_tokenizer")
from paddlenlp.transformers import CLIPVisionConfig
from ppdiffusers.pipelines.stable_diffusion import StableDiffusionSafetyChecker
vision_config = CLIPVisionConfig(**{
    "hidden_size": 32,
    "image_size": 32,
    "model_type": "clip_vision_model",
    "num_attention_heads": 4,
    "num_channels": 3,
    "num_hidden_layers": 5,
    "patch_size": 2,
    "return_dict": True,
  })
safety_checker = StableDiffusionSafetyChecker(vision_config)
# unet
from ppdiffusers import UNet2DConditionModel
unet = UNet2DConditionModel(**{
  "attention_head_dim": 8,
  "block_out_channels": [
    32,
    64
  ],
  "cross_attention_dim": 32,
  "down_block_types": [
    "DownBlock2D",
    "CrossAttnDownBlock2D"
  ],
  "in_channels": 4,
  "layers_per_block": 2,
  "out_channels": 4,
  "sample_size": 32,
  "up_block_types": [
    "CrossAttnUpBlock2D",
    "UpBlock2D"
  ]
})
# vae 
from ppdiffusers import AutoencoderKL
vae = AutoencoderKL(**{
  "act_fn": "silu",
  "block_out_channels": [
    32,
    64
  ],
  "down_block_types": [
    "DownEncoderBlock2D",
    "DownEncoderBlock2D"
  ],
  "in_channels": 3,
  "latent_channels": 4,
  "layers_per_block": 1,
  "norm_num_groups": 32,
  "out_channels": 3,
  "sample_size": 32,
  "up_block_types": [
    "UpDecoderBlock2D",
    "UpDecoderBlock2D"
  ]
})
from ppdiffusers import StableDiffusionPipeline
from ppdiffusers import StableDiffusionPipeline
pipe = StableDiffusionPipeline(
    vae=vae,
    text_encoder=text_encoder,
    tokenizer=tokenizer,
    unet=unet,
    scheduler=scheduler,
    safety_checker=safety_checker,
    feature_extractor=feature_extractor
)
pipe.save_to_hf_hub("tiny-random-stable-diffusion-pipe")

https://huggingface.co/sijunhe/tiny-random-stable-diffusion-pipe/tree/main

[paddle-bot]

Thanks for your contribution!

[codecov]

Codecov Report

Merging #4640 (6edb3b4) into develop (68156b6) will decrease coverage by 0.02%.
The diff coverage is 16.21%.

@@             Coverage Diff             @@
##           develop    #4640      +/-   ##
===========================================
- Coverage    42.11%   42.10%   -0.02%     
===========================================
  Files          433      433              
  Lines        61821    61849      +28     
===========================================
+ Hits         26038    26041       +3     
- Misses       35783    35808      +25     

Impacted Files	Coverage Δ
paddlenlp/transformers/model_utils.py	77.32% <0.00%> (-0.44%)	⬇️
paddlenlp/transformers/tokenizer_utils_base.py	78.14% <0.00%> (-0.24%)	⬇️
paddlenlp/transformers/image_processing_utils.py	24.41% <20.83%> (-1.59%)	⬇️
...lenlp/ops/fast_transformer/transformer/decoding.py	7.72% <100.00%> (ø)

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

[JunnYu]

LGTM